Journal / Kirill Naumov
AI in Retail Training: Practise the Moment, Then Coach the Shift
A practical way to use AI in frontline learning: rehearse one real customer situation, practise on the shop floor and let managers coach—not score—the person.
By Kirill Naumov ·
A colleague can finish an AI role-play and still freeze when a customer asks the same question beside a crowded shelf. That is not a reason to abandon practice. It is a reminder that training has to travel from the exercise into the shift.
Design around one observable behaviour
Choose a moment that matters: asking one useful question before recommending a product, explaining a delivery delay, or handing a complaint to the right colleague. Define what a good response looks like in plain language. Avoid vague goals such as “be more confident” or “sell better”; those are hard to practise and easy to turn into unfair scores.
AI can play a customer with different needs, create safe variations and give the learner another attempt without making a real customer the practice environment. Keep the scenario grounded in approved product facts and service policy. If the answer depends on information the system cannot verify, the practice should reward the learner for checking or asking for help—not for inventing a fluent response.
Connect practice to the real workplace
A 2025 study of 483 retail-industry participants in the [Journal of Retailing and Consumer Services](https://doi.org/10.1016/j.jretconser.2025.104259) examined knowledge transfer in retail training. It is survey evidence, not proof that a particular AI tool improves performance. Still, it points to a question worth asking: does the workplace make it possible to use what people learned?
Build a small transfer loop. Before the exercise, the manager names the behaviour and explains when it is useful. The colleague practises two or three variations with AI. During the shift, the manager observes one appropriate moment and gives brief, specific feedback. Later, the colleague reflects on what changed. The aim is a better work habit, not a transcript archive.
A 2026 working paper reports a quasi-experiment involving nearly 2,000 salespeople at one US industrial firm and an average performance lift of about 2% associated with AI role-play. It is a preprint, in a different industry and selling context, and its authors emphasise the role of transfer into real selling. Treat it as an interesting early signal, not a retail benchmark or a promise of results. [Read the working paper](https://ssrn.com/abstract=5717822).
Keep coaching separate from surveillance
Do not turn every practice attempt into an employee ranking. Explain what is stored, who can see it and how long it is kept. Give staff a way to correct a mistaken assessment and keep the manager accountable for any consequential judgement. Do not record customer interactions by default or infer a colleague’s attitude from voice, face or response speed.
Measure the learning loop at three levels: whether the learner can explain the behaviour, whether they can use it in an observed scenario, and whether the process helps the customer without adding avoidable work. Track manager time, factual errors, escalations and colleague confidence alongside any commercial measure. Compare with the current training method; if AI adds friction or no useful practice, change or stop the pilot.
The goal is not to have a machine grade the people who work in a shop. It is to give them more chances to rehearse a difficult moment and help their manager coach the part that matters. For practical AI applications across retail, sales, learning and loyalty, see the [AI for Retail workshop](https://www.kirillnaumovai.com/workshops/ai-retail-november).